Adaptation

Ad^tation was previously defined as (a) determining what’s different between a new problem and a retrieved case, and then, (b) modifying the solution stored in the retrieved case to take these differences into account. While this definition is taken fiom hterature on Case-based reasoning (Riesbeck and Schank, 1989) it is sufficiently general to cover instances of psychological adaptation in the context of analogical problem solving, where it is the analogically derived solution plan which is subject to modification. However, for the purposes of the present work only (b) is treated as referring to adaptation, as it is only under (b) that alterations are made to a retrieved case. As Section 1.5.1 indicated (a) can be treated as the result of the process of verification. The remainder of this section identifies the issues related to adaptation that this thesis is concerned with.

Adaptation Operations

^goao^lisatioQ’ can be treated as an instance of schema induction (Gick and Holyoak, 1983) and so is best considered in the context of learning from successful analogizing, addition and substitution stand as two candidate operaticxis. The term "candidate’ is used advisedly for two reasons. Firstly, Novick and Holyoak’s investigation was concerned with analogical problem solving in the context of maths problons, ratho- than natural language problems which typically feature in analogical problem solving research. Second^, these operations were not tested fcs* directly, rather t h ^ were identified on the basis of a post hoc categorisation of participants’ answers, both correct answers and errors, to the problems.

These cmsiderations do not prevent Novick and Holyoak’s findings from providing a starting point for further research into adaptation, but they do not entail that any operations they have identified also feature in the context of natural language analogizing.

None-the-less, cm the basis of Novick and Holyoak’s research it is hypothesised that the operations of adding and substituting elements are applied in the context of natural language analogues. However, literature on Case-based reasoning (Riesbeck and Schank, 1989; Kolodner, 1993) suggests that a further operation can be added to those identified by Novick and Holyoak: that of deleting elements. By incorporating deletion into a taxonomy operations apphed to adapt a possible solution plan I advance the following hypothesis: the operations of deleting and adding elements stand as the basic, or primitive, operations, and the operation of substituting elements is the product of a combination of these two basic operations.

Whether these operations are actually applied wfren dealing with natural language problems is examined in Chapter Four, as are the conditions under which they are apphed. Yet before examining any further issues considered in the present work it is important to introduce a conceptual distinction between two possible types of adaptation, types which are not distinguished by their constituent operations, but by the domain to which they are apphed. This is important because this thesis is only concerned with examining one of these types.

Types o f Adaptation: Procedural and Implementational

The type of ad^taticxi referred to by Novidc and Holyoak (1991) can be termed ‘procedural adaptation’ as it is concerned with adapting procedures in a solution plan which has failed to verify as vahd. A further type of adaptation can be termed ‘implementational adaptation’ this refers to adaptation that

Keane (1994) has lecoitfy initiated research into implementational adaptation: This amounts to either the addition of elements to the target domain which have hitherto gone unmentioned but which are required for a solution plan to be implemented, or to the deletion of elements that are judged irrelevant to the inq)Iemaitatioo of a solution plan. It is important to note the use of the term ‘irrelevant’ rather than ai^ term suggesting the deletion of constraints which block the ^plication of a solution plan, as it is in virtue of just such blocking constraints that procedural adaptation is effected in the first place.

An example of implementational adaptation is as follows: generating the convergence solution to Duncker’s radiatim problem involves directing multiple low intensity rays simultaneously at a tumour. Givm that it is initially assumed that there is only m e available source fi’om which to direct the rays, the individual, in solving a problem, must then generate several further sources and add these to their representation of the target problem, in order to implement the convergence solution coherently. Consequently, I suggested that implementational adaptation constitutes a fiirther type of adaption - although reahsed in terms of the same operations as procedural adaptatim - which needs to be incorporated into any complete account of ad^tim , and by extension analogical problem solving itself. However, while acknowledging the distinction between the two types of adaptation the thesis is concerned only with procedural adaptation.

Ease o f Applying Different Adaptation Operations

Apart from estabhshing what adaptation consists in a further issue I cmsider is wfiether there are differences between the ease of applying one adaptation operation rather than another. Novick and Holyoak (1991) determined that the ^h catio n of the operation of adding elements as responsible for the greatest number of errors, compared with the application of the other adaptation operations. Consequently, I hypothesise that f l y i n g the operation of adding elements is relatively more difficult than carrying m t analogical mapping requiring no adaptation (here after no adaptation’). This issue is examined in Chapter Four. I further hypothesise that applying the operatim of deleting elements should be relatively easier to apply than the operation of adding elements. This issue is examined in Chapters Four and Five.

1 further cmtend that differences in the ease of applying the operatims of deleting and adding elements results in a tendency to apply the operation of deleting rather than adding elements to adapt solution plan. This tendency, and the difference in the ease of applying the operations is explained as the result

target demain: specifically to map the least amount of information necessary to realise the goal target problem’s goal state. This aim to minimise effort is characterised as a constraint on individuals analogical problem solving performance. Furthermore, I argue that, as a result of the aim to expend minimal effort, the individual is alreacty disposed to apply the operation of deleting elements, hence its relative ease of application. This issue is examined in Chapter Six.